Searcharxiv⌕ Search

arXiv subjects

Yuan Liu

Publications and source records attributed to Yuan Liu.

At least 361 records · Page 20Linked to original sources
arXiv2023

F$^{2}$-NeRF: Fast Neural Radiance Field Training with Free Camera Trajectories

This paper presents a novel grid-based NeRF called F2-NeRF (Fast-Free-NeRF) for novel view synthesis, which enables arbitrary input camera trajectories and only costs a few minutes for training. Existing fast grid-based NeRF training frameworks, like Instant-NGP, Plenoxels, DVGO, or TensoRF, are mainly designed for bounded scenes and rely on space warping to handle unbounded scenes. Existing two widely-used space-warping methods are only designed for the forward-facing trajectory or the 360-degree object-centric trajectory but cannot process arbitrary trajectories. In this paper, we delve deep into the mechanism of space warping to handle unbounded scenes. Based on our analysis, we further propose a novel space-warping method called perspective warping, which allows us to handle arbitrary trajectories in the grid-based NeRF framework. Extensive experiments demonstrate that F2-NeRF is able to use the same perspective warping to render high-quality images on two standard datasets and a new free trajectory dataset collected by us. Project page: https://totoro97.github.io/projects/f2-nerf.

cs.CV↗
arXiv2023

PixMIM: Rethinking Pixel Reconstruction in Masked Image Modeling

Masked Image Modeling (MIM) has achieved promising progress with the advent of Masked Autoencoders (MAE) and BEiT. However, subsequent works have complicated the framework with new auxiliary tasks or extra pre-trained models, inevitably increasing computational overhead. This paper undertakes a fundamental analysis of MIM from the perspective of pixel reconstruction, which examines the input image patches and reconstruction target, and highlights two critical but previously overlooked bottlenecks. Based on this analysis, we propose a remarkably simple and effective method, {\ourmethod}, that entails two strategies: 1) filtering the high-frequency components from the reconstruction target to de-emphasize the network's focus on texture-rich details and 2) adopting a conservative data transform strategy to alleviate the problem of missing foreground in MIM training. {\ourmethod} can be easily integrated into most existing pixel-based MIM approaches (\ie, using raw images as reconstruction target) with negligible additional computation. Without bells and whistles, our method consistently improves three MIM approaches, MAE, ConvMAE, and LSMAE, across various downstream tasks. We believe this effective plug-and-play method will serve as a strong baseline for self-supervised learning and provide insights for future improvements of the MIM framework. Code and models are available at \url{https://github.com/open-mmlab/mmselfsup/tree/dev-1.x/configs/selfsup/pixmim}.

cs.CV↗
arXiv2023

Client Selection for Federated Bayesian Learning

Distributed Stein Variational Gradient Descent (DSVGD) is a non-parametric distributed learning framework for federated Bayesian learning, where multiple clients jointly train a machine learning model by communicating a number of non-random and interacting particles with the server. Since communication resources are limited, selecting the clients with most informative local learning updates can improve the model convergence and communication efficiency. In this paper, we propose two selection schemes for DSVGD based on Kernelized Stein Discrepancy (KSD) and Hilbert Inner Product (HIP). We derive the upper bound on the decrease of the global free energy per iteration for both schemes, which is then minimized to speed up the model convergence. We evaluate and compare our schemes with conventional schemes in terms of model accuracy, convergence speed, and stability using various learning tasks and datasets.

cs.LG↗
arXiv2023

On the holomorphic convexity of reductive Galois coverings over compact Kähler surfaces

This article generalizes the result of Katzarkov and Ramachandran from algebraic surfaces to Kähler surfaces. We follow their argument to prove the holomorphic convexity of a reductive Galois covering over a compact Kähler surface which does not have two ends, except that we replace the $p$-adic factorization theorem by an analysis of the singularities of the continuous subanalytic plurisubharmonic exhaustion function.

math.CV↗
arXiv2023

The Long-term Monitoring Results of Insight-HXMT in the First 4 Yr Galactic Plane Scanning Survey

The first X-ray source catalog of Insight-HXMT Galactic Plane (|b|<10deg) Scanning Survey (GPSS) is presented based on the data accumulated from June 2017 to August 2021. The 4 yr limit sensitivities at main energy bands can reach 8.2x10^(-12) erg/s/cm^2} (2-6 keV), 4.21x10^(-11) erg/s/cm^2 (7-40 keV) and 2.78x10^(-11) erg/s/cm^2 (25-100 keV). More than 1300 sources have been monitored at a wide band (1$-$100\,keV), of which 223 sources have a signal-to-noise ratio greater than 5. We combined the GPSS data of Insight-HXMT and MAXI and found it is feasible to obtain more complete long-term light curves from their scanning results. The flux variabilities at different energy bands of the 223 bright sources are analyzed based on the excess variances. It is found that the fluxes of X-ray binaries are more active than those of supernova remnants and isolated pulsars. Different types of binaries, e.g., low-mass X-ray binaries (LMXBs), high-mass X-ray binaries (HMXBs), neutron star binaries, and black hole binaries, also distinctively show different regularities. In addition, the relations between the hardness ratio (HR) and excess variances, and HR and source types are analyzed. It is obvious that the HRs of HMXBs tend to be harder than those of LMXBs and HMXBs tend to be more active than those of LMXBs.

astro-ph.HE↗
arXiv2023

Diagnosing failures of fairness transfer across distribution shift in real-world medical settings

Diagnosing and mitigating changes in model fairness under distribution shift is an important component of the safe deployment of machine learning in healthcare settings. Importantly, the success of any mitigation strategy strongly depends on the structure of the shift. Despite this, there has been little discussion of how to empirically assess the structure of a distribution shift that one is encountering in practice. In this work, we adopt a causal framing to motivate conditional independence tests as a key tool for characterizing distribution shifts. Using our approach in two medical applications, we show that this knowledge can help diagnose failures of fairness transfer, including cases where real-world shifts are more complex than is often assumed in the literature. Based on these results, we discuss potential remedies at each step of the machine learning pipeline.

cs.LG↗
arXiv2023

NeuS: Learning Neural Implicit Surfaces by Volume Rendering for Multi-view Reconstruction

We present a novel neural surface reconstruction method, called NeuS, for reconstructing objects and scenes with high fidelity from 2D image inputs. Existing neural surface reconstruction approaches, such as DVR and IDR, require foreground mask as supervision, easily get trapped in local minima, and therefore struggle with the reconstruction of objects with severe self-occlusion or thin structures. Meanwhile, recent neural methods for novel view synthesis, such as NeRF and its variants, use volume rendering to produce a neural scene representation with robustness of optimization, even for highly complex objects. However, extracting high-quality surfaces from this learned implicit representation is difficult because there are not sufficient surface constraints in the representation. In NeuS, we propose to represent a surface as the zero-level set of a signed distance function (SDF) and develop a new volume rendering method to train a neural SDF representation. We observe that the conventional volume rendering method causes inherent geometric errors (i.e. bias) for surface reconstruction, and therefore propose a new formulation that is free of bias in the first order of approximation, thus leading to more accurate surface reconstruction even without the mask supervision. Experiments on the DTU dataset and the BlendedMVS dataset show that NeuS outperforms the state-of-the-arts in high-quality surface reconstruction, especially for objects and scenes with complex structures and self-occlusion.

cs.CV↗
arXiv2023

Orientation-dependent electron-phonon coupling in interfacial superconductors LaAlO3/KTaO3

The emergent superconductivity at the LaAlO3/KTaO3 interfaces exhibits a mysterious dependence on the KTaO3 crystallographic orientations. Here we show, by soft X-ray angle-resolved photoemission spectroscopy, that the interfacial superconductivity is contributed by mobile electrons with unexpected quasi-three-dimensional character, beyond the "two-dimensional electron gas" scenario in describing oxide interfaces. At differently-oriented interfaces, the quasi-three-dimensional electron gas ubiquitously exists and spatially overlaps with the small q Fuchs-Kliewer surface phonons. Intriguingly, electrons and the Fuchs-Kliewer phonons couple with different strengths depending on the interfacial orientations, and the stronger coupling correlates with the higher superconducting transition temperature. Our results provide a natural explanation for the orientation-dependent superconductivity, and the first evidence that interfacial orientations can affect electron-phonon coupling strength over several nanometers, which may have profound implications for the applications of oxide interfaces in general.

cond-mat.supr-con↗
arXiv2023

Intrinsic Bayesian Optimisation on Complex Constrained Domain

Motivated by the success of Bayesian optimisation algorithms in the Euclidean space, we propose a novel approach to construct Intrinsic Bayesian optimisation (In-BO) on manifolds with a primary focus on complex constrained domains or irregular-shaped spaces arising as submanifolds of R2, R3 and beyond. Data may be collected in a spatial domain but restricted to a complex or intricately structured region corresponding to a geographic feature, such as lakes. Traditional Bayesian Optimisation (Tra-BO) defined with a Radial basis function (RBF) kernel cannot accommodate these complex constrained conditions. The In-BO uses the Sparse Intrinsic Gaussian Processes (SIn-GP) surrogate model to take into account the geometric structure of the manifold. SInGPs are constructed using the heat kernel of the manifold which is estimated as the transition density of the Brownian Motion on manifolds. The efficiency of In-BO is demonstrated through simulation studies on a U-shaped domain, a Bitten torus, and a real dataset from the Aral sea. Its performance is compared to that of traditional BO, which is defined in Euclidean space.

stat.ML↗
arXiv2023

Gen6D: Generalizable Model-Free 6-DoF Object Pose Estimation from RGB Images

In this paper, we present a generalizable model-free 6-DoF object pose estimator called Gen6D. Existing generalizable pose estimators either need high-quality object models or require additional depth maps or object masks in test time, which significantly limits their application scope. In contrast, our pose estimator only requires some posed images of the unseen object and is able to accurately predict the poses of the object in arbitrary environments. Gen6D consists of an object detector, a viewpoint selector and a pose refiner, all of which do not require the 3D object model and can generalize to unseen objects. Experiments show that Gen6D achieves state-of-the-art results on two model-free datasets: the MOPED dataset and a new GenMOP dataset collected by us. In addition, on the LINEMOD dataset, Gen6D achieves competitive results compared with instance-specific pose estimators. Project page: https://liuyuan-pal.github.io/Gen6D/.

cs.CV↗
arXiv2023

Error Correction of Quantum Algorithms: Arbitrarily Accurate Recovery Of Noisy Quantum Signal Processing

The intrinsic probabilistic nature of quantum systems makes error correction or mitigation indispensable for quantum computation. While current error-correcting strategies focus on correcting errors in quantum states or quantum gates, these fine-grained error-correction methods can incur significant overhead for quantum algorithms of increasing complexity. We present a first step in achieving error correction at the level of quantum algorithms by combining a unified perspective on modern quantum algorithms via quantum signal processing (QSP). An error model of under- or over-rotation of the signal processing operator parameterized by $ε< 1$ is introduced. It is shown that while Pauli $Z$-errors are not recoverable without additional resources, Pauli $X$ and $Y$ errors can be arbitrarily suppressed by coherently appending a noisy `recovery QSP.' Furthermore, it is found that a recovery QSP of length $O(2^k c^{k^2} d)$ is sufficient to correct any length-$d$ QSP with $c$ unique phases to $k^{th}$-order in error $ε$. Allowing an additional assumption, a lower bound of $Ω(cd)$ is shown, which is tight for $k = 1$, on the length of the recovery sequence. Our algorithmic-level error correction method is applied to Grover's fixed-point search algorithm as a demonstration.

quant-ph↗
arXiv2023

Efficient Fully-Coherent Quantum Signal Processing Algorithms for Real-Time Dynamics Simulation

Simulating the unitary dynamics of a quantum system is a fundamental problem of quantum mechanics, in which quantum computers are believed to have significant advantage over their classical counterparts. One prominent such instance is the simulation of electronic dynamics, which plays an essential role in chemical reactions, non-equilibrium dynamics, and material design. These systems are often time-dependent, which requires that the corresponding simulation algorithm can be successfully concatenated with itself over different time intervals to reproduce the overall coherent quantum dynamics of the system. In this paper, we quantify such simulation algorithms by a property called fully-coherent: the algorithm succeeds with arbitrarily high success probability $1-δ$, while only requiring a single copy of the initial state. We subsequently develop fully-coherent simulation algorithms based on quantum signal processing (QSP), including a novel algorithm that circumvents the use of amplitude amplification while also achieving a query complexity additive in time $t$, $\ln(1/δ)$, and $\ln(1/ε)$ for error tolerance $ε$: $Θ\big( \|\mathcal{H}\| |t| + \ln(1/ε) + \ln(1/δ)\big)$. Furthermore, we numerically analyze these algorithms by applying them to the simulation of the spin dynamics of the Heisenberg model and the correlated electronic dynamics of an H$_2$ molecule. Since any electronic Hamiltonian can be mapped to a spin Hamiltonian, our algorithm can efficiently simulate time-dependent ab initio electronic dynamics in the circuit model of quantum computation. Accordingly, it is also our hope that the present work serves a bridge between QSP-based quantum algorithms and chemical dynamics, stimulating a cross-fertilization between these exciting fields.

quant-ph↗
arXiv2022

Presentations of Galois groups of maximal extensions with restricted ramification

Motivated by the work of Lubotzky, we use Galois cohomology to study the difference between the number of generators and the minimal number of relations in a presentation of the Galois group $G_S(k)$ of the maximal extension of a global field $k$ that is unramified outside a finite set $S$ of places, as $k$ varies among a certain family of extensions of a fixed global field $Q$. We prove a generalized version of the global Euler-Poincaré Characteristic, and define a group $B_S(k,A)$, for each finite simple $G_S(k)$-module $A$, to generalize the work of Koch about the pro-$\ell$ completion of $G_S(k)$ to study the whole group $G_S(k)$. In the setting of the nonabelian Cohen-Lenstra heuristics, we prove that the objects studied by the Liu--Wood--Zureick-Brown conjecture are always achievable by the random group that is constructed in the definition the probability measure in the conjecture.

math.NT↗
arXiv2022

MoQuad: Motion-focused Quadruple Construction for Video Contrastive Learning

Learning effective motion features is an essential pursuit of video representation learning. This paper presents a simple yet effective sample construction strategy to boost the learning of motion features in video contrastive learning. The proposed method, dubbed Motion-focused Quadruple Construction (MoQuad), augments the instance discrimination by meticulously disturbing the appearance and motion of both the positive and negative samples to create a quadruple for each video instance, such that the model is encouraged to exploit motion information. Unlike recent approaches that create extra auxiliary tasks for learning motion features or apply explicit temporal modelling, our method keeps the simple and clean contrastive learning paradigm (i.e.,SimCLR) without multi-task learning or extra modelling. In addition, we design two extra training strategies by analyzing initial MoQuad experiments. By simply applying MoQuad to SimCLR, extensive experiments show that we achieve superior performance on downstream tasks compared to the state of the arts. Notably, on the UCF-101 action recognition task, we achieve 93.7% accuracy after pre-training the model on Kinetics-400 for only 200 epochs, surpassing various previous methods

cs.CV↗
arXiv2022

Non-abelian Cohen--Lenstra Heuristics in the presence of roots of unity

For a Galois extension $K/\mathbb{F}_q(t)$ of Galois group $Γ$ with $\gcd(q,|Γ|)=1$, we define an invariant $ω_K$, and show that it determines the Weil pairing of the curve corresponding to $K$ and it descends to the prime-to-$|Γ|$-torsion part of the lifting invariants of Hurwitz schemes introduced by Ellenberg--Venkatesh--Westerland and Wood. By keeping track of the image of $ω_K$, we compute, as $K$ varies and $q\to \infty$, the average number of surjections from the Galois group of maximal unramified extension of $K$ to $H$, for any $Γ$-group $H$ whose order is prime to $q|Γ|$. Motivated by this result, we modify the conjecture of Wood, Zureick-Brown and the author about non-abelian Cohen--Lenstra, for both function fields and number fields, to cover the cases when the base field contains extra roots of unity. We also discuss how to use the invariant $ω_K$ to construct a random group model, and prove in a special case that the model produces the same moments as our function field result.

math.NT↗
arXiv2022

On the $p$-rank of class groups of $p$-extensions

We prove a local-global principle for the embedding problems of global fields with restricted ramification. By this local-global principle, for a global field $k$, we use only the local information to give a presentation of the maximal pro-$p$ Galois group of $k$ with restricted ramification, when some Galois cohomological conditions are satisfied. For a Galois $p$-extension $K/k$, we use our presentation result for $k$ to study the structure of pro-$p$ Galois groups of $K$. Then for $k=\mathbb{Q}$ and $k=\mathbb{F}_q(t)$ with $p\nmid q$, we give upper and lower bounds for the rank of $p$-torsion group of the class group of $K$, and these bounds depend only on the structure of the Galois group and the inertia subgroups of $K/k$. Finally, we study the $p$-rank of class groups of cyclic $p$-extensions of $\mathbb{Q}$ and the $2$-rank of class groups of multiquadratic extensions of $\mathbb{Q}$, for a fixed ramification type.

math.NT↗
arXiv2022

Modeling Global Distribution for Federated Learning with Label Distribution Skew

Federated learning achieves joint training of deep models by connecting decentralized data sources, which can significantly mitigate the risk of privacy leakage. However, in a more general case, the distributions of labels among clients are different, called ``label distribution skew''. Directly applying conventional federated learning without consideration of label distribution skew issue significantly hurts the performance of the global model. To this end, we propose a novel federated learning method, named FedMGD, to alleviate the performance degradation caused by the label distribution skew issue. It introduces a global Generative Adversarial Network to model the global data distribution without access to local datasets, so the global model can be trained using the global information of data distribution without privacy leakage. The experimental results demonstrate that our proposed method significantly outperforms the state-of-the-art on several public benchmarks. Code is available at \url{https://github.com/Sheng-T/FedMGD}.

cs.LG↗
arXiv2022

Enhancing Cyber Resilience of Networked Microgrids using Vertical Federated Reinforcement Learning

This paper presents a novel federated reinforcement learning (Fed-RL) methodology to enhance the cyber resiliency of networked microgrids. We formulate a resilient reinforcement learning (RL) training setup which (a) generates episodic trajectories injecting adversarial actions at primary control reference signals of the grid forming (GFM) inverters and (b) trains the RL agents (or controllers) to alleviate the impact of the injected adversaries. To circumvent data-sharing issues and concerns for proprietary privacy in multi-party-owned networked grids, we bring in the aspects of federated machine learning and propose a novel Fed-RL algorithm to train the RL agents. To this end, the conventional horizontal Fed-RL approaches using decoupled independent environments fail to capture the coupled dynamics in a networked microgrid, which leads us to propose a multi-agent vertically federated variation of actor-critic algorithms, namely federated soft actor-critic (FedSAC) algorithm. We created a customized simulation setup encapsulating microgrid dynamics in the GridLAB-D/HELICS co-simulation platform compatible with the OpenAI Gym interface for training RL agents. Finally, the proposed methodology is validated with numerical examples of modified IEEE 123-bus benchmark test systems consisting of three coupled microgrids.

cs.LG↗